The CESM Procedure
FORECAST Statement
FORECAST variable-list / <options>;
The FORECAST statement lists the numeric variables in the input data table whose accumulated values represent time series to be modeled and forecast. (The input data table is specified in the DATA= option in the PROC CESM statement.)
You must specify a variable-list that contains one or more numeric variables. For more information about the variable-list, see the section "SAS Variable Lists" in SAS Programmers Guide: Essentials.
You can specify any number of FORECAST statements, but you cannot specify the same variable in more than one of them.
You can specify the following options to specify which forecast model to use:
- ACCUMULATE=option
specifies how to accumulate the data table observations within each time period for the variables in the variable-list. If the ACCUMULATE= option is not specified in the FORECAST statement, accumulation is determined by the ACCUMULATE= option in the ID statement. Use the ACCUMULATE= option with multiple FORECAST statements when you want different accumulation specifications for different variables. For more information, see the ACCUMULATE= option in the ID statement.
- ALPHA=number
specifies the significance level to use in computing the confidence limits of the forecast, where number must be between 0 and 1. By default, ALPHA=0.05, which produces 95% confidence intervals.
- BACK=n
specifies the number of observations before the end of the data where the multistep forecasts are to begin. By default, BACK=0.
- CRITERION=criterion
-
specifies the model selection criterion (statistic of fit) to use to select from several candidate models. The default is CRITERION=RMSE. The following statistics of fit are provided:
- SSE
sum of square error
- MSE
mean squared error
- RMSE
root mean squared error
- UMSE
unbiased mean squared error
- URMSE
unbiased root mean squared error
- MAXPE
maximum percent error
- MINPE
minimum percent error
- MPE
mean percent error
- MAPE
mean absolute percent error
- MDAPE
median percent error
- GMAPE
geometric mean percent error
- MAPES
mean absolute error percent of standard deviation
- MDAPES
median absolute error percent of standard deviation
- GMAPES
geometric mean absolute error percent of standard deviation
- MINPPE
minimum predictive percent error
- MAXPPE
maximum predictive percent error
- MPPE
mean predictive percent error
- MAPPE
symmetric mean absolute predictive percent error
- MDAPPE
median predictive percent error
- GMAPPE
geometric mean predictive percent error
- MINSPE
minimum symmetric percent error
- MAXSPE
maximum symmetric percent error
- MSPE
mean symmetric percent error
- SMAPE
symmetric mean absolute percent error
- MDASPE
median symmetric percent error
- GMASPE
geometric mean symmetric percent error
- MINRE
minimum relative error
- MAXRE
maximum relative error
- MRE
mean relative error
- MRAE
mean relative absolute error
- MDRAE
median relative absolute error
- GMRAE
geometric mean relative absolute error
- MAXERR
maximum error
- MINERR
minimum error
- ME
mean error
- MAE
mean absolute error
- MASE
mean absolute scaled error
- RSQUARE
R-square
- ADJRSQ
adjusted R-square
- AADJRSQ
Amemiya’s adjusted R-square
- RWRSQ
random walk R-square
- AIC
Akaike information criterion
- AICC
Akaike information corrected criterion
- SBC
Schwarz Bayesian information criterion
- APC
Amemiya’s prediction criterion
- LEAD=n
-
specifies the number of periods ahead to forecast (the forecast lead or horizon).
The value n is not relative to the last nonmissing observation of a particular series, but is instead relative to the BACK= option specification to the last observation in the input data table or the accumulated series. Thus, if a series has missing values at the end, the actual number of forecasts computed for that series is greater than n.
By default, LEAD=0.
- MEDIAN
estimates the median forecast values and uses those values for forecasting. (By default, PROC CESM uses mean values for forecasting.) If you do not specify the TRANSFORM= option, no transformation is applied to the time series, so the mean and median forecast values are identical.
- METHOD=model-name
-
specifies the forecasting model to use to forecast the time series. You can specify the following forecasting model-names:
- ADDWINTERS
requests the Winters additive method.
- BEST
requests the best candidate smoothing model among the SIMPLE, LINEAR, DAMPTREND, SEASONAL, ADDWINTERS, or WINTERS methods.
- BESTN
requests the best candidate nonseasonal smoothing model among the SIMPLE, LINEAR, or DAMPTREND methods.
- BESTS
requests the best candidate seasonal smoothing model among the SEASONAL, ADDWINTERS, or WINTERS methods.
- DAMPTREND
requests damped trend exponential smoothing.
- DOUBLE
requests second-order exponential smoothing.
- LINEAR
requests linear (Holt) exponential smoothing.
- MULTSEASONAL
requests multiplicative seasonal exponential smoothing.
- SEASONAL
requests additive seasonal exponential smoothing.
- SIMPLE
requests simple (single) exponential smoothing.
- WINTERS
requests Winters multiplicative method.
By default, METHOD=BEST.
- SETMISSING=option | number
specifies how to assign missing values (either input or accumulated) in the accumulated time series for variables in the variable-list. If the SETMISSING= option is not specified in the FORECAST statement, missing values are set according to the value of the SETMISSING= option in the ID statement. For more information, see the SETMISSING= option in the ID statement.
- TRANSFORM=option
-
specifies the time series transformation to be applied to the input or accumulated time series. You can specify the following values for option:
- AUTO
automatically chooses between NONE and LOG on the basis of the model selection criteria.
- BOXCOX(n)
performs Box-Cox transformation with parameter number (n), where n must be between –5 and 5.
- LOG
performs logarithmic transformation.
- LOGISTIC
performs logistic transformation.
- NONE
performs no transformation.
- SQRT
performs square-root transformation.
By default, TRANSFORM=NONE.
When the TRANSFORM= option is specified, the time series must be strictly positive. After the time series is transformed, the model parameters are estimated by using the transformed series. The forecasts of the transformed series are then computed, and finally the transformed series forecasts are inverse-transformed. The inverse transform produces either mean or median forecasts depending on whether the MEDIAN option is specified. For more information, see the sections Transformations and Inverse Transformations.